Applied tools
Riccardo Rebonato, Alexander Denev · Cambridge University Press eBooks · 2013
A word of caution Bayesian nets are a mature technology and many numerical techniques have been developed to handle them effectively. There are literally dozens of papers to deal with virtually any of the problems that one finds when one begins to use large Bayesian nets in earnest – say, the NP -hardness of the computational burden, the exponential explosion (in the number of parents) of the entries of the conditional probability tables, what to do with underspecified nets, etc. Some of the solutions are frankly quite complex. For reasons we explain below, we are lucky, because most of this sophistication is not required for our applications. So, a few simple tricks can be profitably adopted for our applications. These ‘tricks’ tend to be the simpler ones and they will make our life easier. And these simple, entry-level, techniques have the largest marginal benefit with respect to a totally naive approach to constructing Bayesian nets. This is great, and reassuring. However, it is very important not to get carried away even by this rather limited toolkit. The reason is that, if one becomes overenthusiastic, one risks losing one of the most positive features of our approach: its intuitional appeal, its transparency, and what we call its auditability (by which we mean the ability of a non-technical intelligent person to question its inputs, its outputs and everything in between).